Sampling Methods | 科研方法论

Full Text

Knowledge of sampling methods is essential to design quality research. Critical questions are provided to help researchers choose a sampling method. This article reviews probability and non-probability sampling methods, lists and defines specific sampling techniques, and provides pros and cons for consideration. In addition, issues related to sampling methods are described to highlight potential problems.

In research, the population is the complete set of individuals, events, or objects that exhibit the behaviors and/or possess the characteristics of interest to the researcher (Elfil & Negida, 2017; Omair, 2014). For example, imagine a nurse researcher who is interested in the perceptions of first-time breastfeeding mothers in the United States. Given the sheer size of this population, the researcher could not possibly access all of these women in a study. Instead, the researcher would need to devise a strategy to identify a representative subgroup of first-time breastfeeding mothers from the population. This subgroup is called a sample, and the process of selecting this subgroup from the population is the sampling method (Shorten & Moorley, 2014). The sampling method should be as rigorous as possible to ensure minimum error and bias and to enhance maximum representativeness (Tyrer & Heyman, 2016).

Sampling methods are categorized into probability or non-probability methods (Omair, 2014; Tyrer & Heyman, 2016). Probability sampling methods incorporate an aspect of random selection, which ensures that each case in the population has an equal likelihood of being selected (Shorten & Moorley, 2014). Common types of probability methods include random sampling, systematic sampling, stratified sampling, and cluster sampling. Non-probability sampling methods use an approach in which the sample is selected based on the subjective judgment of the researcher instead of using random selection (Elfil & Negida, 2017). Common types of non-probability sampling methods include quota sampling, purposive sampling, self-selection sampling, and snowball sampling.

Choosing a sampling method requires a researcher to consider multiple factors (e.g., the research question, the study methodology, knowledge about the population of interest, the size of the population of interest, the degree of similarity or differences for cases in the population, and time and/or financial constraints) and the degree of confidence desired for study conclusions along with generalizability (Elfil & Negida, 2017; Shorten & Moorley, 2014). Table 1 presents critical questions for researchers to consider in selecting a sampling method. When the researcher is interested in a population that is relatively small, it may be possible for the researcher to include all persons in the population in a study. When this situation occurs, the study is called a census study. In truth, however, most populations are too large to sample completely. The sampling method assists the researcher in selecting a representative sample and provides guidance on how large the sample needs to be to ensure the degree of confidence desired for conclusions and generalizability.

Table 1. Critical Questions to Help Researchers Choose the Appropriate Sampling Method.

Note. Table developed based on review from Elfil and Negida (2017), Shorten and Moorley (2014), Tyrer and Heyman (2016), and van Hoeven et al. (2015).

Quantitative research questions lend to either probability or non-probability sampling methods, whereas qualitative research questions are limited to non-probability sampling methods. Tables 2 and 3 define each of the common types of probability and non-probability methods and consider pros and cons for each type of sampling method.

Table 2. Definitions of Probability Sampling Methods with Associated Pros and Cons.

Note. Table developed based on review from Elfil &Negida (2017), Shorten & Moorley (2014), Tyrer & Heyman (2016), and van Hoeven et al., (2015).

Table 3. Definitions of Non-Probability Sampling Methods with Associated Pros and Cons.

Note. Table developed based on review from Elfil &Negida (2017), Shorten & Moorley (2014), Tyrer & Heyman (2016), and van Hoeven et al. (2015).

Some Concerns About Sampling Methods

There is an assumption among many that probability sampling methods are always more accurate in capturing the essential characteristics of a population, but that is not always the case. For probability sampling methods, the size of the population should be known (or closely estimated) and persons in the population should be entered into a sampling frame. Probability sampling methods work best when the population of interest is large and when variations from the true population parameters can be identified. Variations from true population parameters are called sampling errors, and these variations can best be identified when sufficiently large samples are chosen.

One key expectation in published randomized controlled trials (RCTs) is to include how the sample size was estimated. Interestingly, a recent article (Tam et al., 2020) reviewed 223 articles with results from RCTs in 80 nursing journals to determine what information was included about sample size estimation. Each article was reviewed to determine if all required components were included to estimate the sample size, and if the value for sample size was computed correctly. Of the 223 articles, 80 (35.6%) did not mention sample size estimation at all. Of the 133 articles that did mention sample size estimation, only 22 (9.9%) included all required components and 35 (41.7%) had estimated sample size values that were incorrect.

A serious challenge for researchers is how to choose sampling strategies for hard-to-reach or hidden populations (e.g., drug users, persons who engage in high-risk sexual behaviors, or persons who are homeless; Firchow & Mac Ginty, 2020; Magnani et al., 2005). For these populations, probability sampling methods cannot be used because members of the population are unknown and there is no sampling frame. Another difficulty is that hard-to-reach or hidden populations represent a relatively small proportion of the general population, thus, statistical interpretations are often unreliable due to small samples and frequent sampling errors. Yet another difficulty is that persons who are members of hard-to-reach or hidden populations are often reluctant to participate in research due to concerns about being stigmatized or fear of negative consequences.

Over the last 20–30 years, the most common sampling strategy for these populations has been snowball sampling (Magnani et al., 2005). While snowball sampling can be a useful non-probability sampling strategy, its effectiveness is often based in the selection of the initial participants and the degree to which those participants are theoretically random. In recent years, there has been a proliferation of new sampling strategies to reach hidden populations with varied success. Firchow and MacGinty (2020) reported great success reaching individuals in conflict-affected societies in low-income countries in Africa with mobile phone surveys and text message surveys. In another recent study, Peven et al. (2019) compared women’s perceptions of breastfeeding support and practices across 11 low-income countries in Africa, Asia, Eastern Europe, and the Caribbean in a secondary analysis from 11 different Department of Health Surveys (DHS) conducted during 2015. Comparisons were made by selecting questions about breastfeeding support and practices that were used in all DHS surveys, which resulted in an overall sample of 87, 328 women. Logistic regressions were performed to examine the potential contribution of approximately 27 predictors to three different models of breastfeeding. While the researchers’ interpretations of the findings suggested some differences by country, it was noted that distinctions in cultural variation and health care policies could not be adequately addressed in the sampling approach.

Conclusion

A basic knowledge of sampling methods is essential to designing quality research. The current article presented probability and non-probability sampling methods, listed and described common types, and presented pros and cons related to their use. Specialized uses for sampling strategies remain complex and should be studied further to identify optimal strategies that will lead to better understanding of unique and complex populations.

References

Elfil M., Negida A. (2017). Sampling methods in clinical research: An educational review. Emergency, 5 (1), Article e52, 1–3.

THE END

打赏