Sampling Methods in Quantitative Research

What are sampling methods in quantitative research

You open your proposal and see that you must specify the survey population, sampling method, sample size, and data collection procedure. These four parts are connected, but they are not the same concept. The sampling method answers one question: how will you select survey units from the population?

The population may consist of university students, customers who have purchased through an e-commerce platform, employees at businesses, or users of an application. The sample is the group of people who actually answer your questionnaire. Since you rarely have enough time and resources to survey the entire population, you need a sample with clear criteria and a documented access process.

In quantitative research, sampling methods are usually divided into two groups. Probability sampling gives each unit in the population a known probability of selection. Non-probability sampling relies on convenience, judgment, quotas, or the researcher's access to participants. Your choice should match the population list, available time, budget, and the level of generalization you need from the results.

You can also review related topics in the Survey category, then compare them with what is sample size to keep the sampling method separate from the number of observations.

How to determine the sampling method and sample size

Before choosing a sampling approach, write one sentence describing the research population. The sentence should include the target group, screening conditions, location or access channel, and survey period. For example: “The research population consists of people aged 18 or older who purchased cosmetics through an e-commerce platform in Vietnam during the most recent 12 months.” The more specific this sentence is, the easier it is to check whether someone belongs in the sample.

If you have a complete population list, probability sampling has a stronger basis. Simple random sampling requires a numbered list and a random selection tool. Systematic sampling can select every kth unit after you determine the sampling interval. Stratified sampling is suitable when the population contains groups that need representation, such as geographic regions or age groups. You can read the detailed procedure in random sampling.

If you do not have a population list, a common option is convenience sampling with screening conditions. You distribute the questionnaire through channels that the target group actually uses, then keep only responses that meet the criteria. This saves time, but it usually provides weaker representation. State this limitation clearly in the methodology chapter instead of calling the data a random sample.

For sample size, Hair et al. propose 5 to 10 observations for each item in some quantitative studies (Hair et al., 2010). If the questionnaire contains 30 items, this rule gives a reference range of 150 to 300 valid responses. This is a planning rule, not a promise that every model will meet its requirements with one fixed number.

If you run a regression with m independent variables, a commonly cited rule is n at or above 50 + 8m (Tabachnick and Fidell, 2013). With 6 independent variables, the reference level is 98 observations. When the study includes several comparison groups, many latent variables, or an expected number of rejected responses, you should collect more than the minimum.

For a finite population with a known population size N, the Yamane formula is often used in Vietnamese theses (Yamane, 1967):

n = N / [1 + N(e²)]

Here, n is the sample size, N is the population size, and e is the allowed error. State the value of e that you selected instead of copying only the final result. If you are unsure which formula to use, compare the approach with sample size calculation and what is the sample size of a data sample.

The table above is for planning. In Chapter 3, you also need to state where the sample was recruited, during which period, what criteria were used to retain responses, and how many valid responses remained after cleaning. If you write only “250 samples were selected using convenience sampling” without describing the access channel, the committee may immediately ask who the sample represents.

Designing the questionnaire for this objective

The sampling method and questionnaire must fit together. Before writing the items, create a table with the research objective, construct to be measured, item codes, scale, and screening conditions. This helps you avoid collecting many responses while still lacking data for one of your hypotheses.

Place screening questions at the beginning of the questionnaire. For example, you might ask whether the respondent has used the service during the most recent 6 months, which age group they belong to, or which area they currently live in. If a respondent does not meet the criteria, the form should end or move to a thank-you message. Do not allow someone outside the population to continue and then remove the response manually at the end of the file.

Latent variables are usually measured with multiple items. For attitude or perception scales, a 5-point or 7-point Likert scale is familiar. Likert introduced a technique for measuring attitudes with ordered response levels (Likert, 1932). State both endpoints clearly, such as 1 for “Strongly disagree” and 5 for “Strongly agree.”

The wording must suit the actual respondents. One item should express one idea, without a sentence that is too long or contains two different actions. If you have reverse-coded items, mark the item code in the coding sheet and check it again before running Cronbach's Alpha. A reverse-coded item handled incorrectly can reduce correlations between items and lead you to remove the wrong item.

The questionnaire should include a short introduction, the survey purpose, estimated completion time, data-use principles, and necessary contact information. Do not add many demographic items if they will not be used to describe the sample or compare groups. Every additional item increases completion time and the risk that a participant will abandon the questionnaire.

You should pilot the questionnaire with a small group that fits the research population to identify unclear wording, missing response options, or incorrect page logic. This checks the instrument, so do not report it as an official result of the main study. After making revisions, lock the questionnaire before collecting the main dataset so that different groups do not answer different versions.

Conducting data collection

Plan data collection in three parts: where you will recruit participants, how you will check eligibility, and when the survey will be open. The access channel must have a reasonable connection to the population. If the study concerns people who have purchased through a particular platform, approach communities or lists likely to contain that group instead of distributing the questionnaire broadly and assuming that every respondent qualifies.

Give each channel a separate tracking code if you want to know where responses came from. For example, you might create a link or code field for university students, office employees, and former customers. Do not use tracking codes to collect identifying information when the research does not require it.

DoThesis has fillform.info, our own form tool for creating questionnaires and collecting responses in Vietnamese. You can also use Google Forms if your university or research group already works with it. Whichever platform you use, check the exported .csv or .xlsx file, column names, Likert coding, and the way the form handles unanswered items before distributing it officially.

Record the collection period using specific dates, such as the start date and end date. While the survey is open, track the number of valid responses rather than only the number of form visits. If one group begins to account for too large a share, you can pause that channel and approach an underrepresented group, provided that you record the adjustment.

You should also plan for invalid responses. If you need 250 valid responses, sending exactly 250 invitations creates a high risk of falling short. The number of invitations required depends on the channel, questionnaire length, and fit between recipients and the target population. Record separately the number collected, the number rejected, and the final number of observations entered into SPSS.

Cleaning data before importing it into SPSS

After closing the survey, keep one copy of the original dataset and create a separate copy for cleaning. Do not edit the original directly. In the working copy, use short variable names without diacritics or spaces, and keep naming consistent across scale groups, such as PU1, PU2, SAT1, and SAT2.

First, remove rows that do not complete the questionnaire or fail the screening questions. Check rows with the same response selected for almost every item. This may be straight-lining, but review the completion time and response pattern before removing the row, since some people may genuinely choose the same level for several statements.

Next, check values outside the permitted range. A 1 to 5 Likert scale must not contain 0, 6, or free text in the analysis columns. “Do not know” or “not applicable” responses need separate codes and must be handled according to the original design. Do not automatically recode them as 3, because that creates a response that the participant never gave.

Check missing data by row and by item. If one respondent leaves too many items blank, remove the case according to the criterion stated in the methodology. If only a few items are missing, the treatment depends on the software and analysis plan. The important point is to decide the criterion before looking at the hypothesis results, so that you do not remove data simply because doing so makes the model look better.

Check reverse-coded items before running Cronbach's Alpha. For a 1 to 5 scale, the reverse score is often coded with the formula 6 - original score, but apply it only to the item that was designed as reverse-coded. Then compare descriptive statistics, frequencies, and variable codes in SPSS. The article sample size in SPSS can help you check how the software displays the number of observations.

Finally, keep a cleaning log with the row code, reason for removal, processing date, and person who performed the action. You do not need to place the entire log in Chapter 4, but keep it available in case your supervisor asks why the number of responses fell from the initial collection to the number used in the analysis.

How to present this in Chapter 3

The sampling method section in Chapter 3 should answer these questions in order: who is the population, what is the unit of analysis, what are the selection criteria, how is the sample selected, where is the data collected, during which period, and what is the expected sample size. If you use convenience sampling, use the correct name and give the practical reason, such as the absence of a complete population list or restricted access.

You can use the following structure. Start by describing the population and survey participants. Then describe the sampling method, questionnaire distribution channels, and screening questions. Next, explain how the sample size was determined, how many responses were collected, how many were rejected, and how many valid responses remained. Finish by describing the instrument, Likert scale, and data-entry procedure.

How to write this in your thesis

“ The study used [probability/non-probability] sampling, specifically [method name], to reach [research population] through [location/survey channel] from [date] to [date]. Based on [rule or formula] (Hair et al., 2010), the study expected to collect [number] responses and used [number] valid responses after data cleaning.”

If you use the rule of 5 to 10 observations for each item, count the items actually entered into the model. Do not count personal-information questions when they are not analysis variables. If you use the Yamane formula, replace N and e with the actual values from your study. Do not copy a sample size from another thesis simply because the subject area is the same.

Common sampling errors

The first error is using the phrase “random sampling” for every online survey. A publicly posted link does not automatically create a random sample. If you do not have a sampling frame and do not know the selection probability of each unit, describe the method honestly as non-probability sampling.

The second error is confusing the sampling method with the questionnaire distribution channel. Distribution through Facebook, Zalo, or email describes access. You still need to state who was eligible to respond, which criteria were used to retain responses, and how you controlled the target group.

The third error is reporting only the final number of responses. Readers need to know the initial number collected, the number removed, and the reasons for removal. If cases were removed because of missing answers, straight-lining, or failed screening, state the criteria before presenting the results.

The fourth error is selecting the sample size before finalizing the model. When you add independent variables, comparison groups, or scales, the required sample size may change. Lock the item list and preliminary analysis plan before setting the collection target.

The fifth error is collecting responses mainly from one easy-to-reach group. For example, the study may concern consumers in general, but 80% of responses may come from your classmates. You may still use the data if the design permits it, but you must describe the sample characteristics and limits on generalization. Do not call the sample representative of all consumers.

Frequently asked questions

Which sampling method is appropriate?

There is no single method that fits every study. If you have a population list and need stronger generalizability, consider probability sampling. If you do not have a sampling frame, convenience sampling with screening conditions may be feasible, but you need to describe its limitations clearly.

What is the sampling method if the survey uses Google Forms?

Google Forms is a tool for creating and receiving questionnaires, not a sampling method. The method depends on how you identify the people to invite, such as convenience sampling, quota sampling, or sampling from an existing list.

Is convenience sampling acceptable in a thesis?

It can be acceptable in many student studies if you describe its nature accurately, explain why you selected it, set screening criteria, and acknowledge the limitations. Do not rename the method as random sampling to make the methodology appear more rigorous.

What is the minimum sample size for a sampling method?

Sample size depends on the number of items, number of independent variables, type of analysis, and population structure. Common rules include 5 to 10 observations for each item (Hair et al., 2010) or n at or above 50 + 8m for regression (Tabachnick and Fidell, 2013). State the rule you selected and show the calculation using your study's actual data.

What should I do if data are missing or many rows contain similar answers?

Create a copy of the dataset, check each case against the criteria you set in advance, and record the reason for removal. Do not delete rows in bulk simply because one variable has an unattractive distribution. After cleaning, report the initial number of observations, the number removed, and the final number entered into SPSS.

Open your questionnaire file and write a paragraph now describing the population, screening criteria, sampling method, sample-size formula, and data-cleaning plan using your study's actual numbers. Then use M4 data analysis to run the analysis and check your .sav or .csv file.