Print

Elements of Sampling Plans

The following will present information on the elements of sampling plans in both qualitative and quantitative research, a part of a work unit in the Track 2 Dissertation Research Seminar courseroom.

Objectives

This document will discuss the elements of the sampling plan, which include the:

• Sampling strategy.
• Sampling design.
• Size of the sample.
• Method for determining the size.
• Recruitment plan.

This document will also help ensure that you are able to use the first three in writing your own sampling plan for your dissertation topic.

There are Many Types of Probability Sampling

There are basically two sampling strategies available:

• Probability Sampling, which is synonymous with random sampling. But sampling can be divided into two steps:
  • Selection.
  • Assignment.

  We'll talk later about assignment.

  Probability sampling describes the selection portion of that process. Probability sampling means that all the members of the target population have an equal chance to participate in your study.

• The other sampling strategy is Non-probability Sampling, which is all the other strategies which are not random. Put simply, non-probability sampling means that everyone in the population does not have an equal chance to participate in your study. Some will have a higher chance of being selected than others. It is all about the odds.

Main Designs within Probability Sampling

You need to be aware that, in Track 2:

• We don't expect you to have mastered research design! So, although some of the following information is about the varieties of probability and non-probability sampling, we don't expect you to work expertly this level of detail in Track 2. But we do want you to be aware that as you approach Track 3, your comps exam, and the dissertation,
• You will need to find primary sources on sampling within your methodology and become knowledgeable about the options, so that you can make an educated decision about which strategy and design is best for your study. So, start getting yourself acquainted now. Before we talk about the types, or as we say, designs of probability and non-probability sampling,
• You need to know the term: Sampling Frame. It's not complicated. The sampling frame is the subset of the larger population from which you'll actually select your participants. For example, if you were evaluating client satisfaction with treatment at a particular agency, you could get a list of all clients in the past twelve months—or however long you want to consider. That list does not have all the clients of the agency, but it is the subset you will select participants from. Got it?

Okay. Let's look at the next element of a sampling plan, the sampling design.

Sampling Designs within Probability Sampling Strategy

Three of the main designs of probability or random sampling are:

• Simple random sampling. In this case, you simply figure out some way to randomly pick names from your sampling frame. Remember the client list of the clients who attended a clinic in the past year? That would be the sampling frame. You might assign a serial number to each name, then generate a list of random numbers and select the names that correspond to your random number list. Simple.
• Stratified random sampling. If there are meaningful subgroups within the population, you might choose this type. Here, you divide the population into homogeneous groups, then take a random sample from each group. For example, if the population were all the clients of the clinic and within that population African Americans and Latino clients were a small minority, you might use stratified random sampling to ensure that you got a fair representation of each minority group into your sample. You'd divide the whole client list in groups by race or ethnicity, then randomly sample within each homogeneous group. A simple random sample might miss them.
• Cluster random sampling. If your study involves a widely dispersed population, such as all the clients in a state-wide group of clinics, you might consider cluster random sampling. Here, you would decide on some logical or meaningful way to divide the population into clusters. For instance, you might learn that the clinics are clustered in cities of a certain size, so you'd make a list of all the cities of that size in the regions served by the company and then randomly select a number of the cities—and therefore of the clinics. Finally, you would collect your data from every client in each clinic.
• Other random sampling methods. There are quite a few other refinements on the themes above, such as systematic random sampling, multi-stage sampling, and others. William Trochim, from whom much of this presentation has been adapted, points out that people who perform applied social science research typically use more sophisticated sampling techniques than these. This is why we want you to realize that more complex techniques exist and that you will need to explore them when you finally design your dissertation.

Reference

Trochim, W. M. K. (n.d.). Probability sampling. Research Methods Knowledge Base. Retrieved from http://www.socialresearchmethods.net/kb/sampprob.php

Sampling Designs within Non-Probability Sampling Strategy

Purposive sampling. This kind of sampling has a purpose in mind. There is something specific about potential participants that you want to know about, or something about them that makes them good informants on your question. The key is that not everyone in a wide population is likely to have the necessary experience or information. For instance, if the study is about depression, you might purposively sample for people who are or have been depressed. In another study, if only 30-32 year old female elementary school teachers could give you information about your study topic, you would purposively seek them. There are some important subcategories of the purposive sampling design. The first is:

• Modal instance or typical instance sampling. You sample for the most typical members of the population. It's hard to know what a typical person is, no matter what the population is, so this is a fairly weak sampling design.
• Expert sampling. This is stronger. If you want to know something about therapists or executive leaders of small organizations, you purposively sample for exactly those persons, because they will have the expert knowledge you're seeking. Or if you want to know about the experience of something, you purposively sample for people who have had significant experience with that.
• Quota sampling. Suppose your population has some specific and important characteristics that occur in some proportion or percentage in the population. It might be meaningful to select that same proportion or percentage in your sample. There are two versions, proportional and non-proportional.
• Proportional quota sampling. In a study of persons suffering from a particular illness, the population is characterized as 30 percent male and 70 percent female. Proportional quota sampling would set those numbers as the quota of males and females it will select. In other words, if the sample size will be 100, the researcher would deliberately select 30 males and 70 females, and would continue sampling until she reached both figures.
• Non-proportional quota sampling. Rather than specifying an exact percentage or proportion, you specify a minimal number that you believe is fairly representative of each group. Thus, in our previous example, we might specify only that we'll have at least 20 males and at least 50 females, assuming that this would fairly represent the population.
• Heterogeneity sampling, or diversity sampling. If the population is diverse, and you're interested in obtaining a representative array of opinions or attitudes within that population, selecting a broad and diverse range of participants who are more likely to have all those attitudes and opinions.
• Snowball sampling. This means that after you have an initial group of participants, you expand the sample by asking your participants to recommend others who might be added to the sample. This is usually a fallback kind of sampling because it is quite likely not to result in a sample that fairly represents the population. It also has some IRB challenges, such as the possibility of undue pressure on new recruits to volunteer.

The final, and least representative kind of non-probability sampling is called:

• Convenience sampling. This is "I'll take whoever I can get" sampling. A common example is the use of college freshmen in Psychology 101 as participants in a professor's study. They are convenient. Asking for volunteers from the general population or from a specific target population is convenience sampling. If you have any hope of generalizing your findings to the population, convenience sampling makes that impossible.

Sample Sizes

Like most issues we discuss in Track 2, sample size can be fairly complex, but we look at the basics here. When you get close to your comprehensive examination and when you take your method courses, look more deeply into the issue of sample sizes. But let's look at those basics.

• Quantitative sample sizes are calculated statistically. Various textbooks suggest different rules for calculating sample sizes, and in experimental or quasi-experimental designs, one typically uses a power analysis to calculate the sample size. A power analysis is a kind of statistical calculation based on the effect size, the power desired in the hypothesis test, and level of statistical significance to be used. Do not rely on a textbook that gives one-size-fits-all estimates. If you read, "Generally, 30 subjects will be sufficient for this study," go to a primary source instead and do a power analysis or find a survey sampling article to calculate a more accurate sample size.
• Qualitative sample sizes are not calculated directly. Many authors recommend different "ballpark" sizes, but here's a basic method to follow in estimating your qualitative sample size. It is not required nor essential, but it gives beginning qualitative researchers a rule of thumb that can help. First,
1. Step 1: find eight or 10 published articles using your design: (that is, case study, ethnography, grounded theory, and so on). What are their sample sizes? What seems to be a fair average or trend in those articles? Set that as a baseline. Next, reflect on the
2. Step 2: Diversity of the target population:
• If the population contains a wide range of information, on your topic, your sample should be larger;
• If a very narrow range of information, the sample can be smaller. Now, if you decide the sample should be relatively larger, increase the baseline number a reasonable amount. Do the opposite for a smaller sample from a homogeneous population.
3. Step 3: Now, consider the design itself. If it is going to require a fairly deep and richly detailed data set, then fewer participants may be appropriate to keep the analysis manageable. Let's look at a few examples.

Applying the Method for Selecting Sample Size in Qualitative Designs

We'll look at sample sizes in three common qualitative designs, applying the three steps we learned in the last slide. Our three examples will be grounded theory, phenomenology, and generic qualitative inquiry.

Step 1. A recent survey of nine Grounded theory articles: showed an Average sample size of 23. Similarly, a recent survey of 11 Phenomenology articles: showed an Average sample size of nine. Finally, a survey of 12 Generic qualitative inquiry articles: revealed an Average sample size of 27.

We'll set these averages as our baseline:

• Grounded theory 23.
• Phenomenology 9.
• Generic qualitative 27.

Estimating Sample Size in Grounded Theory

Here's our baseline: Grounded theory 23.

Step 2: What Range of diversity is in our target population? For this example, let's assume the study is investigating how young adult African American women college graduates described their process of adjusting to and thriving in the college environment. A little side research reveals that for a number of reasons, this population— female young adult African American college graduates—is fairly homogeneous. Since we can estimate that there won't be a huge variability in their experiences, we can keep to our average or baseline number. But we go on to Step 3: How deep and nuanced will our information have to be? Probably fairly deep, and we hope nuanced, This study won't be more nuanced or deeper than the usual grounded theory study, though, and grounded theory always seeks rich and nuanced information, so it is reasonable to stick with our baseline. So let's set our Minimum sample size at 23.

Let's look at setting sample size for a phenomenological study.

Estimating Sample Size in Phenomenology

Here's our baseline from step 1: Phenomenology 9.

Step 2: What Range of diversity is in our target population? For this example, let's assume the study is about the lived experience of feeling jealous. It's not particularly reasonable to assume that jealousy is experienced widely differently across people, and certainly the clinical literature of its cousin, paranoia, suggests that paranoid thinking is fairly uniform across populations. So there seems to be no particular reason of diversity to increase from our baseline of nine. Next, we go on to Step 3: How deep and nuanced will our information have to be? Phenomenology always goes deeply into the participants' lived experience. Hundreds of pages of data are normal. So this argues against raising the sample size, but because the average number of participants is nine, then nine must be a reasonably manageable number. So let's set our Minimum sample size at 9.

Finally, let's see what happens with generic qualitative inquiry.

Estimating Sample Size in Generic Qualitative Inquiry

Here's our baseline from step 1: Generic qualitative 27.

Step 2: What range of diversity is in our target population? For this example, let's assume the study is asking elementary school teachers their opinions on the value to them of belonging to a teachers' union. This is clearly a very diverse population who will have a wide range of opinions on a very sensitive and important topic. Is 27 participants going to be enough to fairly represent the opinions in such a diverse group? Probably not. We might want to increase our sample size to 45 or 50, and probably we'll want to use a purposive heterogeneous sampling design. Next, we go on to Step 3: How deep and nuanced will our information have to be? Fortunately, our research question does not require much sophistication and nuance. There will be many different opinions, obviously, but they will equally fall into a few categories. Analysis will not be as deep and textured as in grounded theory or phenomenology. So let's set our Minimum sample size at 50.

Have you noticed that in each instance we have set a minimum sample size? What's that about?

Information Saturation in Qualitative Analysis

We want information saturation. In quantitative sampling, sample size is simply calculated. If a power analysis requires a sample of 40, then you select 40 participants. You don't need more (except to account for attrition). But in qualitative inquiry, what if you reach your sample size but new information has been emerging even in the last few participants? In this case, we say that the study has not reached information saturation, which is the situation in which no new information is emerging in data collection. Some writers call this theoretical saturation, which is the term for it in grounded theory. There is no set rule for determining saturation; the researcher must make a good judgment. Some mentors suggest setting an arbitrary number like this: No new information in the last two interviews in a row. However, there is no standard for this except the judgment of the researcher.

The desire for information saturation requires that the researcher add additional participants if saturation is not met by the stated number. For example, if we say we'll interview nine participants in a phenomenological study, and if after nine interviews new stories and experiences are still being described, we'll need more participants. Therefore, we set a minimum number of participants, leaving ourselves open to recruiting more participants if we need them to achieve saturation. Or, some mentors prefer that we set a range of participants, such as nine to fifteen. But even there, if saturation is not achieved by the upper limit, we continue to recruit until it is.

The final element of the sampling plan is the plan for actually recruiting participants.

The Recruitment Plan and Conclusion

Recruitment plans are part of the sampling plan. The recruitment plan is not something you should concern yourself too much with at Track 2. It is highly dependent on the final shape of the research question and the actual ultimate design. It is very closely linked to the nature of the target population and the actual resources of time, money, and energy the researcher will have at the time of the study. All these will become clearer as you finish Track 3 and start your dissertation, working on the final version of your design with your mentor. So typically, it won't be determined until then. In work unit of the courseroom, you'll have a chance to review an example of a sampling and to play with creating one of your own. But we know and you should be aware that this will, out of necessity be only provisional at this point.

To Repeat: The Five elements of the Sampling Plan are:

• The sampling strategy, which is:
• Probability.
• Non-probability.

·  The sampling design is:

• Random if (probability), or,
• Purposive, snowball, or convenience if (non-probability).

·  And the sample size which is determined by:

• In Quantitative: Power analysis or other calculation.
• In Qualitative: Three-step analysis of:
• Similar studies in the literature (setting a baseline), of
• The degree of diversity in the population (that is, more diverse means larger sample).
• The nature of the design itself (that is, a dense, rich, nuanced analysis, means smaller sample, simpler, analysis means larger sample).

Thank you for your time and attention.

Please return to the courseroom and complete the Self-Assessment: Elements of Sampling Plans.

---

Doc. reference: phd_t2_u08s3_elementssp.html