What Is Data Collection? How to Do It for a Quantitative Thesis

What Is Data Collection

Data collection is the process of obtaining the information you need from a group of respondents and recording it in a structure that is clear enough to analyze. In a quantitative thesis, the data usually sits in a questionnaire containing items, Likert scales, demographic questions, and several screening variables.

The process starts before you open SPSS. You need to know which constructs your topic measures, who should answer the questionnaire, how each variable will be coded, and which column will contain each answer in the .sav or .csv file. If you collect responses from the wrong population or write questions that do not measure the intended construct, running Cronbach's Alpha or EFA later will not fix the underlying problem.

You can view quantitative data as a chain: research model, variables and scales, questionnaire, respondents, raw data, clean data, and analysis results. There is no single formula that applies to every topic. Sample size and collection method depend on the population, number of items, model, and your access to respondents.

You can also review Statistics if you are connecting data collection with Cronbach's Alpha, EFA, regression, or SmartPLS.

Why Data Collection Matters in Quantitative Research

Data collection determines three things in your thesis. First, the data shows whether you can test the hypotheses you proposed. A hypothesis about the effect of service quality on satisfaction requires measurable variables for both constructs, together with responses from the correct customer group.

Second, the data affects the reliability and validity of your scale. A questionnaire with many items can still produce noisy data when respondents do not understand the items, select the same response for every item, or answer despite being outside the target population. You may see a high Cronbach's Alpha, but that alone does not prove that the questionnaire measured the intended construct.

Third, the data determines whether you can explain your procedure. When your supervisor asks where you sampled, during which period, how many valid responses you had, and which responses you removed, your answer must match both the data file and the methodology chapter. Do not write only that you collected enough responses. Save the questionnaire link, survey opening date, screening criteria, and cleaning log.

In quantitative research, data is commonly collected through a structured questionnaire. You can distribute the questionnaire online, send it by email or through community groups, or collect responses directly using an electronic form. The distribution method is only the channel. Methodological quality depends on whether respondents belong to the target population and whether the items were designed consistently.

How Much Data Is Enough

The question of how much data is enough is often treated as though it had one fixed answer. In practice, you need to separate at least four questions: how many responses you received, how many are valid, whether each item has enough data, and whether the sample fits the model.

If you use a rule based on the number of items, Hair et al. suggest considering approximately 5 to 10 observations for each item (Hair et al., 2010). This is a planning basis, not permission to collect any sample and call it representative. If your model has 25 items, this approach gives a reference range of 125 to 250 responses. You still need to consider the number of independent variables and the analysis technique.

If you run a regression with m independent variables, a commonly cited rule is a minimum sample size n of at least 50 + 8m (Tabachnick and Fidell, 2013). The two calculations can produce different results. When they do, select the level that fits your model and your supervisor's requirements, and allow for responses that may be removed.

The table below is illustrative output showing how you can track the collection process. The numbers only illustrate how to record the process. They are not the results of a real study.

You should collect more responses than the expected minimum. The removal rate differs across topics, so do not promise in advance that every response will be retained. In Chapter 3, you can report the number of questionnaires distributed, responses received, removal criteria, and final number of valid responses.

How to Read Collected Data in the Output

After exporting the data, SPSS does not give you a table called data collection quality. You need to inspect several locations. Start with Data View to review response rows and Variable View to check variable names, value labels, data type, Missing, and Measure.

In Variable View, an item such as PU1, PU2, or PU3 may be defined as Numeric, with Decimals set to 0, Values from 1 to 5, and Measure set to Scale. Gender or age-group variables may also be coded with numbers while remaining categorical variables in substance. A column containing numbers does not automatically make the variable Scale.

You can run a frequency table through Analyze > Descriptive Statistics > Frequencies. Select the screening and demographic variables, then review Frequency, Percent, Valid Percent, and Cumulative Percent. If a group has a frequency of 0, check the coding or the way the form was exported.

The table below is illustrative output showing how SPSS commonly displays the results. It is not a finding from any thesis.

The N column tells you how many values were used in the statistics for that variable. Missing tells you how many values are blank. Minimum and Maximum help you detect out-of-range codes. For example, a Likert scale may allow only values from 1 to 5, while one row contains 7. Mean and Std. Deviation give you an initial view of the distribution, but they are not enough to conclude that a scale is good or poor.

You should also check for duplicate responses. If one person submits the form twice, keeping both rows may distort group proportions and make the sample appear larger than it really is. Check the respondent ID if you collected a separate identifier. If no identifier exists, remove a response only when you have evidence such as submission time, response pattern, or duplicated information, and record the decision.

If you see a few highly unusual values during checking, they may be outliers. Do not delete a row simply because the result looks better afterward. Determine whether it is a data-entry error, a valid but extreme response, or an unreliable response.

What to Do When Data Collection Falls Short

If the number of responses is low, first check your sample criteria and access channels. Are you sending the questionnaire to the correct target group, or only to a convenient group? A convenience sample may be usable within the limits of your topic, but you must describe the sampling method accurately and avoid claiming that the sample represents the entire population.

If many respondents leave items blank, check whether the item is required, difficult to understand, or asks for information that respondents may be reluctant to provide. For collected data, you can set a rule for removing responses with too much missing information. Do not fill a large number of blank cells with the mean unless you have a clear methodological reason.

If one response level dominates across variables, check whether the wording is leading and whether respondents have actually experienced what you asked about. For example, a survey about app experience loses meaning when respondents have never used the app. Place screening questions before the scale items so that the wrong respondents are filtered out at the beginning.

If you do not yet have enough valid responses, you can extend the collection period or add another channel, but keep the sampling criteria and questionnaire version unchanged. If you change the wording, number of Likert levels, or order of item groups midway through collection, record when the change occurred. Data from two versions may no longer be fully equivalent.

Some problems cannot be repaired with software. You cannot turn a respondent from the wrong population into a member of the target population by changing a code. You cannot make an item that measures the wrong construct valid by deleting a few rows. If the problem is in the questionnaire design, discuss it with your supervisor early and decide whether to recollect the data or limit your conclusions.

Data Collection Versus Scale Design

Data collection is the act of obtaining responses from respondents. Scale design is the decision about which constructs, items, wording, and response scale you will use to measure those constructs. The two processes are connected, but they do not replace each other.

For example, you may collect 300 responses for a questionnaire about intention to use, but if the items ask only about perceived ease of use, the data does not fully measure intention. Conversely, a well-built scale still does not produce useful data if you send the questionnaire to the wrong respondents.

You also need to distinguish control, mediator, and moderator variables. A moderator changes the strength or direction of the relationship between two variables, while a mediator explains the mechanism behind that relationship. When your model includes a moderator, you need to collect the information required for grouping or interaction testing. You can review how to use moderator and moderating to avoid entering the variable's role incorrectly in the data file.

Finally, data collection is different from data entry. A form may export a file automatically, but you still need to check coding, variable names, missing data, and invalid responses before running the analysis.

Common Errors

The first error is distributing the questionnaire before finalizing the model and scales. You may then lack a variable for one hypothesis or use the same code for two different constructs. Prepare the variable-code list first, such as PU1 to PU4 and SAT1 to SAT4, then compare it with the questionnaire.

The second error is changing the Likert coding between item groups. If 1 means strongly disagree in one section but strongly agree in another, you create inconsistent data. Reverse-coded items must be identified clearly and handled deliberately before running the reliability analysis.

The third error is looking only at the number of responses and ignoring quality. A complete data row can still reflect straight-line responding, an unrealistically short completion time, or failure to pass the screening question. Keep one raw-data file, one cleaned-data file, and a log that records each type of removed response.

The fourth error is removing data based on the test result. You should not delete a group of respondents simply because they reduce Cronbach's Alpha, unless you have an independent methodological criterion that you can explain. Re-running the analysis repeatedly to find a better-looking result increases the risk of difficult questions at the defense.

The final error is reporting numbers in Chapter 3 that do not match Chapter 4. The number of valid responses, survey period, sampling method, and sample characteristics must be checked against the final file used for analysis.

Frequently asked questions

What is data collection in quantitative research?

It is the process of obtaining structured information from research respondents and turning it into numerical data, usually through a questionnaire. The data must connect to the variables, scales, sample criteria, and analysis plan you established.

How much data collection is enough?

There is no single correct number for every topic. You can refer to 5 to 10 observations per item (Hair et al., 2010), or n at least 50 + 8m for a regression with m independent variables (Tabachnick and Fidell, 2013), then compare the result with your model and your supervisor's requirements.

Should you use Google Forms for data collection?

Google Forms can handle a basic questionnaire and export the data. fillform.info is DoThesis's own form tool, suitable when you want to manage the questionnaire and responses within a separate workflow. Whichever tool you use, check the respondents, missing data, duplicate responses, and coding before importing the file into SPSS.

What should you do when data is missing?

First determine whether the missing data appears in a few cells or across entire responses. For a response with too much missing information, you can remove it according to a rule set in advance. For a few missing cells, the appropriate treatment depends on the missing proportion, variable type, and analysis requirements. Record your decision and avoid replacing every blank with the mean without an explanation.

Can you delete responses to improve Cronbach's Alpha?

You should remove a response only when you have an independent criterion, such as the respondent being outside the target population, too much missing data, a duplicate response, or evidence of unreliable responding. Do not remove it only because it lowers Alpha. If you remove an item after running the analysis, state the basis for the decision and update the results table, scale table, and discussion.

How to write this in your thesis

Open your data file now, create a copy of the raw data, prepare the variable-code table, and record the number of responses at each stage before running the statistics. If you need support running the analysis on your own .sav or .csv file and checking the workflow, see DoThesis M4 data analysis.