What Is Data Processing? The SPSS Process from Raw Data to Results

What Is Data Processing

Data processing is the series of tasks that turns initial survey responses into a structured dataset that you can analyze and explain in your thesis. For a quantitative study, you usually start with an Excel file, Google Forms export, or CSV file. You then check missing data, code variables, reverse-code items when necessary, detect unusual responses, and import the file into SPSS to run the appropriate analyses.

You can think of this workflow as having three layers. The first is data cleaning, such as checking blank cells, values outside the Likert scale, and rows with highly similar responses. The second is variable processing, which includes naming variables, assigning value labels, and calculating composite or mean variables. The third is analysis, which includes descriptive statistics, Cronbach's Alpha, EFA, correlations, regression, or model testing.

Data processing is not about making the numbers look better. The goal is to ensure that every number in the output reflects the responses you collected and answers the research question you defined. If you change data only because Alpha is low or the p-value is not significant, the committee may question the result as soon as you defend the thesis.

In SPSS, the work usually begins in two main windows. Data View contains one row for each respondent, while Variable View contains information about variable names, data types, variable labels, value labels, and display settings. A usable file needs a consistent structure in both places.

If the file is open and you are unsure where to begin, you can also review SPSS topics to compare the analyses commonly used in theses.

Why Data Processing Matters in Quantitative Research

Data processing determines which conclusions you can draw from your questionnaire. An item entered with the wrong coding direction can reduce Cronbach's Alpha. A column containing text instead of numbers can prevent SPSS from including the variable in regression. Rows that are removed without being recorded can make the sample size in Chapter 3 different from the sample size shown in the output.

A sound workflow helps you answer these questions in order:

• Does the survey sample contain enough usable information for analysis?
• Are the items in the same scale consistent with one another?
• Do the items form the factors predicted by the theoretical model?
• Can the independent variables explain the dependent variable?
• Which hypotheses are supported by the test results?

The order of analysis matters as well. You should not run regression immediately after importing the file. First, check the data, describe the sample, assess scale reliability, and evaluate convergent validity and discriminant validity if you use SmartPLS. In a two-step approach, you assess the measurement model first and then move to the structural model, consistent with (Anderson and Gerbing, 1988).

With SPSS, the tables you commonly encounter include Frequencies, Descriptives, Reliability Statistics, Item-Total Statistics, KMO and Bartlett's Test, Total Variance Explained, Rotated Component Matrix, Correlations, Model Summary, ANOVA, and Coefficients. Save the output after each run and use clear names such as 01_lam_sach, 02_alpha, and 03_efa, rather than keeping only one final file.

If you are processing Cronbach's Alpha specifically, the article Cronbach Alpha focuses on Cronbach's Alpha, Corrected Item-Total Correlation, and the Cronbach's Alpha if Item Deleted column.

What Level of Data Processing Is Acceptable

There is no single number that applies to the entire workflow. Each statistic answers a different question, so you need to read the correct table and use the correct source. The table below lists thresholds commonly used in quantitative research. These thresholds are not automatic permission to remove an item.

These thresholds need to be interpreted in the context of your study. A new scale or exploratory study may need a different explanation from an established scale. When reporting the results, state the observed value, the criterion used for evaluation, and the supporting citation.

For example, the sentence “The scale's Cronbach's Alpha was 0.812, above the 0.7 threshold, so the scale had good reliability according to (Nunnally, 1978)” is more useful than “The scale met the standard.” This wording tells the reader which statistic you checked and what you used as the basis for the decision.

How to Read Data Processing Results in the Output

Start by identifying the question that the output table answers. Reliability Statistics reports Alpha for the full scale, but it does not yet show which item reduces reliability. To examine that issue, read Item-Total Statistics as well, especially Corrected Item-Total Correlation and Cronbach's Alpha if Item Deleted.

Illustrative output: Cronbach's Alpha

The tables below show illustrative output, not the results of a specific study.

In this example, the overall Alpha is 0.812. DV4 has a Corrected Item-Total Correlation of 0.218, below 0.3. If you delete DV4, Alpha increases to 0.846. You still need to review the content of DV4 before removing it. The item may have been reverse-coded but not reverse-scored, or it may measure a different aspect that the theoretical model intentionally includes.

Illustrative output: EFA

The KMO value of 0.781 is above 0.5, and Bartlett's Test has Sig. 0.000, so the illustrative data are suitable for considering EFA under the criterion in (Kaiser, 1974). Total variance explained is 58.427%, above the commonly used 50% benchmark in (Hair et al., 2010). When reading Rotated Component Matrix, check which factor each item concentrates on, whether the loading is large enough, and whether there is substantial cross-loading.

In practice, A first EFA run that comes out clean is unusual. Removing an item, running the analysis again, and recording each removal round is normal. Keep a tracking table with the item code, the reason for removal, the loading or corrected item-total correlation, and the corresponding output version.

What to Do When the Data Processing Results Do Not Meet the Criteria

The first step is to check for data-entry errors. Open Variable View and check whether the variable is set to Numeric, whether the Likert values fall within the designed range, and whether reverse-coded items have been processed. If the questionnaire uses a scale from 1 to 5, a value of 55 is usually a coding error rather than an unusual observation.

The second step is to check incomplete responses. Decide in advance what qualifies as a row for removal, such as a row with too many blank items or one that does not complete the screening section. Do not delete rows in bulk simply because a few blank cells make the results look better. Record the initial number of observations, the number removed, and the number remaining after cleaning.

The third step is to review items with low statistics. For Alpha, check Corrected Item-Total Correlation and the item content. For EFA, check the loading, cross-loading, and fit with the theoretical construct. An item with a low loading may need to be removed, but the decision needs a methodological reason, not just the desire to increase one number.

The fourth step is to rerun the analysis after each controlled change. Do not remove several items at the same time if you do not know which item caused the problem. Save the output after every round and update your processing log. If removing an item changes the scale structure, compare the result with the model and the hypothesis section again.

Some problems cannot be fixed by software. If the data were collected from the wrong respondents, too many rows follow the same repeated pattern, or an important variable was left out of the questionnaire, report the issue honestly to your supervisor. Running the analysis repeatedly does not turn low-quality data into high-quality data.

If the problem is that items move into different groups in EFA, review the guide to the rotated matrix. If the study requires SEM with AMOS, distinguish between the CB-SEM and PLS-SEM workflows before choosing software. Do not place CFI, TLI, and RMSEA statistics in a SmartPLS report.

Data Processing Compared with Data Analysis

Data processing is broader than running one statistical test. It includes cleaning, coding, standardizing the file structure, creating variables, and preparing the data. Data analysis is the part that uses statistical procedures to answer the research question, such as running regression, EFA, or a test of differences.

For example, changing the gender variable from the text values “Male” and “Female” to the codes 1 and 2 is data processing. Calculating the frequency of the gender variable is descriptive statistics. Comparing the mean scores of two gender groups with an Independent-Samples T Test is data analysis.

You also need to distinguish data processing from data entry. Data entry only puts responses into a structure that the software can read. After entry, you still need to check variable names, missing values, reverse coding, outliers, and the plausibility of each column.

SPSS, Stata, AMOS, and R do not replace this reasoning process. SPSS is convenient because of its menus and the familiar output used in many theses. Stata software is suitable for many econometric analyses and panel data. AMOS is commonly used for CB-SEM. Any tool requires correct input data and clear reporting criteria.

Common Errors

The first error is using accents, spaces, or overly long names in variable names. Use short codes such as DV1, DV2, and YD1, then record the full meaning in the Label field. This makes the output easier to read and reduces errors when selecting variables.

The second error is entering text responses into a variable that needs to be calculated. If the item responses range from “Strongly disagree” to “Strongly agree,” convert them into numerical codes using one consistent rule and record that rule in your processing documentation.

The third error is forgetting to reverse-code an item. For a negatively worded item on a scale from 1 to 5, the usual reverse-coding formula is 6 - original value. Apply this only when the questionnaire and scale design actually require it. Do not reverse-code every item whose wording differs from the others.

The fourth error is removing an item as soon as Alpha increases. A higher Alpha alone does not prove that the scale is better if removing the item eliminates important content. Combine the statistical result with the construct definition, the scale source, and the research model.

The fifth error is copying numbers from the output into the thesis without saving the corresponding data-file version. Name files according to the processing round, keep the original file in read-only mode, and record the date of each change. When your supervisor asks why the sample size decreased or why an item no longer appears in EFA, you can reopen the process instead of guessing.

Frequently asked questions

What is data processing in SPSS?

It is the process of preparing and checking data before, during, and after running analyses in SPSS. The work includes cleaning the file, coding variables, handling reverse-coded items, checking missing data, running tests, and interpreting the output.

How much data processing is acceptable?

There is no single threshold for the entire process. Review each statistic, such as Alpha, Corrected Item-Total Correlation, KMO, Bartlett's Test, total variance explained, or factor loading, and compare it with the appropriate source in your research methods section.

What should I do when data are missing?

First determine whether the missing data appear in a few cells or across an entire row. Then apply the criteria you defined in advance to retain the row, remove it, or handle the missing value. Record the number of changes and the reasons for them. Do not remove rows simply because doing so produces a more convenient result.

Should I delete an item immediately when Cronbach's Alpha is low?

Not yet. Check Corrected Item-Total Correlation, Cronbach's Alpha if Item Deleted, the item content, and whether the item was coded in the reverse direction. If there is a basis for removal, rerun the scale and record which item was removed, why it was removed, and how Alpha changed.

Can I process data in Excel instead of SPSS?

Excel is suitable for preliminary checks, format changes, and detecting some data-entry errors. For analyses such as EFA, regression, or scale testing, SPSS produces a more familiar structured output for a thesis. You can prepare the file in Excel and save it as .xlsx or .csv before importing it into SPSS.

How to write this in your thesis

Open a copy of the data file, check Variable View, create a change log, and then run descriptive statistics, reliability, and EFA in sequence before moving to the main model. If you need to run the analysis on your own .sav or .csv file, see M4 data analysis.