SERVQUAL Model: Dimensions, Scales, and Testing

What is the SERVQUAL model

The SERVQUAL model measures service quality through the gap between customers' expectations and their perceptions after using a service. Parasuraman, Zeithaml, and Berry developed the model, whose original scale uses multiple items to assess service quality from the customer's perspective (Parasuraman et al., 1988).

In a quantitative thesis, you can use SERVQUAL to answer questions such as which factors affect service quality, which factor customers rate lowest, or how service quality affects satisfaction and continued usage intention. The model fits banking, hospitals, schools, hotels, transportation, e-commerce, and many other service settings.

Remember that SERVQUAL measures service quality through respondents' perceptions. You therefore need to define the target respondents, service context, and point at which customers have enough experience to provide an assessment. Do not copy a questionnaire from one industry into another and immediately run Cronbach's Alpha. Adjust the wording of the items, pilot the questionnaire, and test the scale again with your study data.

For a quick overview of the foundations of this model, you can also read SERVQUAL before designing the model for your own thesis.

Dimensions of the model

The classic SERVQUAL model is usually presented through five dimensions. Each dimension represents an aspect customers use to evaluate a service. The table below gives the interpretation commonly used in research in Vietnam, based on the SERVQUAL scale (Parasuraman et al., 1988).

These five dimensions are often treated as independent variables or latent variables in a research model. You can measure service quality as an overall dependent variable, or treat the five dimensions as factors that directly affect satisfaction.

For example, if your topic is “Factors affecting customer satisfaction with digital banking services”, the model may include TAN, REL, RES, ASS, and EMP as independent variables, with SAT as the dependent variable. Each latent variable needs a group of items, such as REL1, REL2, REL3, and REL4.

The abbreviations do not have to match the table above. You can use HH for tangibles, TC for reliability, or DU for responsiveness. What matters is consistency across the questionnaire, .sav file, .csv file, SPSS, and Chapter 4.

The original model and common extensions

The original SERVQUAL model distinguishes between customers' expectations and their perceptions after using a service. Under this logic, service quality is assessed through the gap between what customers expect and what they actually perceive. In some studies, respondents rate the same item twice, once for expectations and once for perceptions.

A two-part questionnaire can become long and increase the chance that respondents skip items or answer by habit. For that reason, many theses measure perceptions only, then adjust the model name and explanation accordingly. State this choice clearly in the methods chapter. Do not claim to have used the full SERVQUAL model when your questionnaire contains only the perception section.

One extension is to adapt the five dimensions to a specific context. For example, a healthcare service study may add cost, accessibility, or professional outcomes when those topics fit the research question. Any new variable still needs a clear definition, theoretical basis, and set of items. Adding a factor simply to increase the number of variables will leave the committee asking for the scale source and the reason for including it.

You may also encounter the SERVPERF model, which measures service quality mainly through perceptions of service performance. SERVPERF is often compared with SERVQUAL when the researcher wants a shorter questionnaire or wants to avoid measuring expectations. Do not combine the two approaches in one scale until you have defined the theoretical model clearly.

You can read about the SERVPERF model to compare measuring perceptions with measuring both expectations and perceptions. If your topic focuses on the sources of quality problems in the delivery process, the five-gap service quality model is more suitable for questions about management gaps.

Another extension connects service quality with satisfaction, perceived value, loyalty, or repurchase intention. In that model, SERVQUAL provides the explanatory variables, while the outcome variables need scales appropriate to their respective concepts. For example, do not use one item to conclude that customers are loyal when your model needs to test a multidimensional latent variable.

Common scales for each concept

Parasuraman et al.'s original SERVQUAL scale is the main reference for the five service quality dimensions (Parasuraman et al., 1988). The table below is not a validated translation for every industry. Adjust the wording to your context, then test it through a pilot study and with the official dataset.

If you use a Likert scale, you can choose 5 or 7 levels ranging from strongly disagree to strongly agree. The Likert scale originated from Likert's technique for measuring attitudes (Likert, 1932). In the questionnaire, write a common instruction that clearly identifies the service being evaluated and the period of experience respondents should use as their reference.

A good item should focus on one idea. The statement “Staff provide fast advice and resolve requests accurately” contains two ideas, speed and accuracy. If respondents rate these two aspects differently, you will not know which idea the score represents. Separate them into two items if both need to be measured.

After data collection, the process usually starts with data screening and descriptive statistics, followed by Cronbach's Alpha, EFA if the study uses SPSS, and then regression or a structural model. Cronbach's Alpha from 0.7 upward is generally considered acceptable (Nunnally, 1978), while a new or exploratory scale may be considered at a level from 0.6 under (Hair et al., 2010). Corrected Item-Total Correlation from 0.3 upward is commonly used to assess the suitability of an item (Nunnally and Bernstein, 1994).

If you run EFA, KMO from 0.5 upward and a statistically significant Bartlett test with a p-value below 0.05 are commonly reported criteria (Kaiser, 1974). Factor loading from 0.5 upward and total variance explained from 50% upward can be considered with reference to (Hair et al., 2010). These are criteria for making a decision, not automatic reasons to delete an item. Consider the theoretical content, cross-loadings, and the effect of deleting the item on the scale structure together.

Applying the model to your topic

Suppose you are studying customer satisfaction with food delivery services in Ho Chi Minh City. The model has five independent variables based on SERVQUAL and one dependent variable, satisfaction. The conceptual model can be expressed through the following hypotheses:

• H1: Tangibles have a positive effect on customer satisfaction.
• H2: Reliability has a positive effect on customer satisfaction.
• H3: Responsiveness has a positive effect on customer satisfaction.
• H4: Assurance has a positive effect on customer satisfaction.
• H5: Empathy has a positive effect on customer satisfaction.

If your proposal requires only four hypotheses, you can select four dimensions based on the literature review and industry characteristics. Explain why you omitted one SERVQUAL dimension. You might argue that it does not fit the research context, has been merged into another concept, or was removed from the model after the scale design stage.

The general regression model is: SAT = β0 + β1TAN + β2REL + β3RES + β4ASS + β5EMP + ε. SAT is satisfaction, β is the effect coefficient, and ε is the error term. The equation helps you visualize the relationship between the variables. The actual results must come from your data file.

When writing the questionnaire, place screening items before the SERVQUAL items. For example, respondents should confirm that they used the service during the period you defined. Only then should you ask about TAN, REL, RES, ASS, EMP, and SAT. Check that the columns in your data file use the correct variable codes, do not contain text in numeric cells, and do not include incomplete responses.

If the study examines a specific brand, replace “service provider” with the brand name in the instruction. If it covers an entire industry, state clearly which type of service respondents are evaluating. A questionnaire that uses a vague phrase such as “this service” may lead respondents to think about several different experiences.

How to test the model with SPSS or SmartPLS

SPSS is suitable when you want to test reliability, EFA, correlations, and linear regression with composite variables. Start with Analyze > Scale > Reliability Analysis, place the items for each dimension into the same run, then select Statistics and tick Scale if item deleted. The Reliability Statistics table reports Cronbach's Alpha, while Item-Total Statistics helps you inspect Corrected Item-Total Correlation and Alpha if Item Deleted. This is the illustrative output described by these tables.

Then run EFA through Analyze > Dimension Reduction > Factor. In Descriptives, select KMO and Bartlett's test. Under Extraction, inspect the eigenvalue and the Total Variance Explained table. Under Rotation, choose a method that fits the research design and inspect the Rotated Component Matrix. If items jump across several groups, check the wording, cross-loadings, and theoretical structure before deleting anything.

For regression, select Analyze > Regression > Linear. Put SAT in the Dependent box and the composite variables TAN, REL, RES, ASS, and EMP in the Independent box. Under Statistics, select Estimates, Model fit, Collinearity diagnostics, and Durbin-Watson. The Coefficients table provides the coefficient, t-statistic, and Sig., while Model Summary provides R Square. Durbin-Watson between 1 and 3 is commonly considered when checking residual autocorrelation (Field, 2013).

SmartPLS is suitable when your model contains several latent variables, requires simultaneous assessment of the measurement model and structural model, or includes mediation or moderation paths. Import the .csv data, create the constructs TAN, REL, RES, ASS, EMP, and SAT, then draw the arrows according to the hypotheses. In SmartPLS 4, run PLS-SEM Algorithm to inspect Outer Loadings, Construct Reliability and Validity, and Discriminant Validity.

For the measurement model, you can report outer loading, Cronbach's Alpha, rho_A, Composite Reliability, and AVE. Outer loading from 0.7 upward is a reference threshold in PLS-SEM (Chin, 1998). CR from 0.7 upward and AVE from 0.5 upward are commonly used to assess composite reliability and convergent validity (Fornell and Larcker, 1981). HTMT should be below 0.85, or below 0.90 for closely related concepts (Henseler et al., 2015).

Next, run Bootstrapping to test the path coefficient, t-value, and p-value. Set 5,000 bootstrap subsamples in line with the procedure commonly cited for PLS-SEM (Hair et al., 2022). For the structural model, you can report VIF, R², f², Q², and path coefficients. VIF below 5 is a reference threshold (Hair et al., 2019), while R² values of 0.75, 0.50, and 0.25 are interpreted as substantial, moderate, and weak in PLS-SEM (Hair et al., 2011). Q² greater than 0 indicates predictive relevance (Stone, 1974).

SPSS has the advantage of a familiar interface and works well for regression, but you must create composite variables yourself and check each step. SmartPLS is convenient for latent variables and models with many paths, but do not include CFI, TLI, and RMSEA from CB-SEM in a PLS-SEM report. Those indices belong to CB-SEM assessment under (Hu and Bentler, 1999).

How to write this in your thesis

You can begin the results paragraph as follows: “The test results show that the [name of dimension] dimension has a [positive/negative] effect on [dependent variable], with β = [value] and p-value = [value]. Because the p-value is [less than/greater than] 0.05, hypothesis H[ ] is [accepted/not supported].” Replace the placeholders with the actual figures from the Coefficients or Path Coefficients table. Do not copy illustrative figures from a tutorial.

Common mistakes when using this model

The first mistake is calling every service quality study SERVQUAL even when the questionnaire does not contain the five dimensions or draw on Parasuraman et al.'s original work. Describe your model accurately. For example, use “a SERVQUAL-based adapted model” if you added, removed, or merged factors.

The second mistake is using an industry's items word for word in another industry. An item about “staff at the counter” does not fit a fully online digital banking application. Adjust the context, pilot the questionnaire, and record the changes.

The third mistake is deleting items only to increase Alpha. A high Alpha does not automatically prove that a scale is valid. An item may increase Alpha while remaining important to the content of the construct. Consider Corrected Item-Total Correlation, item content, loading, and factor structure together.

The fourth mistake is claiming a causal effect from a cross-sectional survey simply because a regression coefficient has a small p-value. In the discussion chapter, use phrases such as “has a statistically significant effect” or “has a relationship under the tested model”, and state the limitations of the research design.

Finally, do not confuse SERVQUAL with the five-gap model. SERVQUAL focuses on measuring customers' perceptions of service quality, while the gap model explains discrepancies in the design and delivery of a service. If you need a foundational definition, read What are the five service quality gaps. Distinguishing the two models in the proposal helps you select the right variables and testing procedure.

Frequently asked questions

Does the SERVQUAL model have to measure both expectations and perceptions?

The original model uses an approach based on expectations and perceptions. However, many adapted studies measure perceptions only to shorten the questionnaire. If you choose this approach, state clearly that it is a SERVQUAL-based adapted model and explain the reason in the methods chapter.

How many dimensions does the SERVQUAL model have?

The common presentation includes five dimensions: tangibles, reliability, responsiveness, assurance, and empathy (Parasuraman et al., 1988). A specific study may adjust, merge, or add dimensions, but every change needs a theoretical basis and must fit the service context.

Should you run SERVQUAL with SPSS or SmartPLS?

If the model mainly uses EFA and regression with composite variables, SPSS is usually sufficient. If you want to assess latent variables, outer loading, CR, AVE, HTMT, and several structural relationships at the same time, SmartPLS is more suitable. Software does not replace scale development or model interpretation.

Why does the Cronbach's Alpha of the SERVQUAL scale fail to meet the threshold?

Check missing data, reverse-coded items, incorrect coding, and items assigned to the wrong group. Review Corrected Item-Total Correlation and the content of each item before deleting anything. Cronbach's Alpha from 0.7 upward is generally accepted (Nunnally, 1978), while an exploratory scale may be considered at a level from 0.6 under (Hair et al., 2010).

Can SERVQUAL be used to measure customer satisfaction?

SERVQUAL mainly measures service quality. Build satisfaction as a separate construct if it is the dependent variable in your model. The five SERVQUAL dimensions can then explain SAT, while SAT needs its own items and an appropriate scale source.

Open your proposal and data file, confirm the five dimensions you are using, check the variable codes, and select the correct SPSS or SmartPLS procedure before the first run. If you need help running the analysis on your own .sav or .csv file, see M4 data analysis by DoThesis.