Attitude Scale: How to Develop and Test It in SPSS

What is an attitude scale, and what does it measure

An attitude scale measures respondents' level of agreement, evaluation, or perception toward an object, behavior, product, or research issue. For example, a study of students' attitudes toward online learning might measure their positive perceptions, willingness to use it, and evaluation of this learning format.

Attitude is usually a latent variable. You cannot observe an attitude directly, so you measure it through multiple observed variables in the questionnaire. Each observed variable is a statement that respondents rate, usually on a 5-point or 7-point Likert scale. The Likert scale was introduced in the work of Likert (1932).

An attitude scale might include statements such as “I have a positive evaluation of this product” or “I am willing to continue using this product.” Respondents choose from 1, strongly disagree, to 5, strongly agree. You can calculate the construct score from the observed variables after the scale meets the requirements for reliability and validity.

You need to distinguish attitude from behavioral intention and actual behavior. “I have a positive attitude toward online shopping” measures attitude, while “I intend to shop online next month” is closer to intention. This distinction prevents you from mixing statements for several constructs in the same scale. If you are building the full model, you can also review research models.

Original scale sources

Do not begin an attitude scale by inventing every item yourself. Look for a scale that has been used for a similar construct, read its original context, and then adapt the wording to your study. A suitable scale source should define the construct clearly, use reasonably consistent observed variables, and come from a context close enough to your research.

Among the permitted sources, foundational models can help you identify attitude in different contexts. Davis (1989) is commonly associated with perceived usefulness and technology acceptance through TAM. Ajzen (1991) treats attitude as a component related to behavioral intention in TPB. For extended technology-acceptance research, UTAUT and UTAUT2 are presented in Venkatesh et al. (2003) and Venkatesh et al. (2012).

The table below is a source-mapping guide. Check the number of observed variables in the original paper and in the version you select. Do not assume that every context uses the complete scale unchanged.

If your study examines service quality, the SERVQUAL scale may be a separate reference direction. You can read more about the SERVQUAL model. If your study examines technology, identify clearly whether you are measuring attitude, perceived usefulness, perceived ease of use, or intention to use. These constructs may be related, but they are not the same.

When citing the source in Chapter 3, state the authors, year, original context, and the way you adapted the scale. If you translated the scale into English or another language, describe it as a translated and adapted version. Do not call it a validated standardized scale unless you have followed an appropriate validation procedure.

Reference set of observed variables

The statements below illustrate the structure of an attitude scale. They are a starting point for designing your questionnaire, not a confirmed translation or a test result. Replace the text in square brackets with the specific object in your study, read every statement again, and check the wording with your supervisor.

Keep one main idea in each observed variable. The statement “I find this service convenient and reliable” contains two ideas. A respondent may agree that it is convenient but disagree that it is reliable. Their response then does not tell you which aspect they were evaluating.

Avoid leading wording, such as “A modern and convenient service like [object] is certainly essential.” This statement suggests the answer and uses several conclusive adjectives. Write briefly, identify the subject, and use words that your survey sample actually understands.

If you use a reverse-coded variable, record it in the data code immediately, for example ATT3_R. When entering the data in SPSS, reverse-code the variable before calculating scores or running Cronbach's Alpha. Adding a reverse-coded statement simply to make the questionnaire appear more rigorous is usually unhelpful. A confusing reverse-coded statement can cause response errors and reduce data quality.

Adapting the scale to the Vietnamese context

Start by comparing the original definition with the research object in Vietnam. If the original source measures attitudes toward a banking application, you cannot immediately use it for attitudes toward a hospital, university, or consumer product by changing a few words. The same label, “attitude,” may involve a different object, experience, and response context.

Review the content and language

Read each observed variable aloud. Check the subject, time frame, object, and how familiar the wording is. “I value the service ecosystem” may suit technology users, but it may confuse respondents who are unfamiliar with the term. Use the same name for the object throughout the questionnaire.

Check for overlapping variables

Place the observed variables next to each other and underline repeated content. ATT1, “positive evaluation,” and ATT5, “positive attitude,” may be too close. Similar ideas are not automatically an error, but if respondents see the same statement with only a few words changed, the scale adds little information.

Pilot the questionnaire

Before the official distribution, ask a small group to read the questionnaire and mark statements that are unclear, open to several interpretations, or unsuitable for real-world use. This checks the wording, but it does not establish that the scale has acceptable reliability. After revising it, save the question version and variable codes so you can compare them with the data file.

When collecting responses, you can use fillform.info, DoThesis's own form tool for Vietnamese questionnaires. Google Forms is a common alternative. Whichever tool you use, export the data as .csv or another format suitable for SPSS, and retain the original file so you can investigate unusual rows later.

Testing scale reliability and validity

The usual sequence is to check the data, run Cronbach's Alpha, run EFA if the research design requires it, and then take the retained variables into regression or the next model. Do not delete a variable simply to make the output look better. Each deletion needs a content-based and statistical reason, and you should record every version of the analysis.

Checking Cronbach's Alpha

In SPSS, go to Analyze > Scale > Reliability Analysis. Move the ATT variables into the Items box and select Alpha as the Model. In Statistics, select Scale if item deleted to view Corrected Item-Total Correlation and Cronbach's Alpha if Item Deleted.

Cronbach's Alpha of 0.7 or above is commonly considered acceptable according to (Nunnally, 1978). For a new scale or exploratory research, Alpha from 0.6 may be considered according to (Hair et al., 2010). A Corrected Item-Total Correlation of 0.3 or above is a commonly used threshold according to (Nunnally and Bernstein, 1994).

Do not read Alpha by itself. If Alpha increases substantially when ATT2 is deleted, check whether ATT2 differs in content, is difficult to understand, or measures another construct. Deleting a variable only because the coefficient increases can remove an important aspect of the scale.

Checking EFA when the study uses EFA

If your model needs to explore the structure of the variables, go to Analyze > Dimension Reduction > Factor. Select the observed variables and check KMO and Bartlett's Test, Total Variance Explained, and Rotated Component Matrix. For rotation, choose a method consistent with your assumption about relationships between factors, then inspect factor loadings and whether one variable loads on several factors.

A KMO of 0.5 or above and a statistically significant Bartlett's test at below 0.05 are used to assess whether the data are suitable for factor analysis according to (Kaiser, 1974). The rule of an eigenvalue greater than 1 is presented in (Kaiser, 1960). Total variance explained of 50% or above and a factor loading of 0.5 or above are commonly referenced thresholds according to (Hair et al., 2010).

These thresholds do not replace reading the item content. If ATT4 has a low loading, review the statement and context before deleting it. If ATT2 loads strongly on another factor, check whether the statement measures another construct or whether respondents understood it differently from what you intended.

Reference criteria table

The table below is illustrative output.

If you use SmartPLS instead of EFA in SPSS, you will read Outer Loadings, Composite Reliability, AVE, and HTMT under the PLS-SEM procedure. Do not place CFI, TLI, or RMSEA criteria from CB-SEM in the SmartPLS assessment. For a study using SPSS, you can also review the AIDA model, the Maslow's hierarchy of needs, or the SWOT model, but choose a model that fits your research question and data.

How to write this in your thesis

In Chapter 3, present the construct definition, scale source, adaptation process, number of observed variables, Likert format, and coding in that order. If you translated the scale from English, describe it as translated and adapted for the research context. The Cronbach's Alpha and EFA results usually belong in Chapter 4, after you have collected and cleaned the data.

You can adapt the following paragraph: “The attitude scale toward [object] was adopted and adapted from [author, year]. The original scale consisted of [number] observed variables and used a [5/7]-point Likert scale, from 1, [description of the lowest level], to [5/7], [description of the highest level]. After reviewing the wording for the context of [object/research sample], the study used variables ATT1 to ATT[number].”

You should also prepare a coding table containing the variable code, observed-variable wording, source, and a note about any reverse-coded variable. Do not write that the scale “meets the requirements” in Chapter 3 before running the data. Chapter 3 describes the planned design. The decision to retain or delete variables must be based on the actual output.

Common mistakes

The first mistake is using “attitude” as a general label and putting statements about perceptions, satisfaction, and intention inside it. Return to the construct definition and classify each statement before creating the variable in SPSS.

The second mistake is translating every word in an English scale. A grammatically correct sentence can still sound unnatural to Vietnamese respondents. Check the subject, context, familiarity, and practical meaning of the sentence.

The third mistake is taking observed variables from several sources without checking whether they fit together. A combined scale may produce a high Alpha while still lacking content validity. Record the source of each variable group and the reason for each adaptation.

The fourth mistake is repeatedly deleting variables until the statistics look good. If someone asks why ATT3 was removed, you must be able to explain its content, item-total correlation, factor loading, or cross-loading. Without a log of each run, it becomes difficult to describe the procedure honestly.

The final mistake is entering Likert codes incorrectly or forgetting to reverse-code a variable. Before running the analysis, use Analyze > Descriptive Statistics > Frequencies to check the minimum, maximum, number of missing values, and codes outside the expected range. A coding error at this stage can distort every later test.

Frequently asked questions

How many observed variables are enough for an attitude scale?

There is no single number that applies to every study. The number depends on the construct definition, scale source, number of aspects that need to be measured, and model design. You need enough variables to represent the construct's content, then test reliability and validity with real data.

Should an attitude scale use a 5-point or 7-point Likert scale?

Both can be used. A 5-point Likert scale is often easier for a general survey sample to answer and interpret, while a 7-point scale provides more distinctions in evaluation. State the meaning of every level, use the same format throughout the questionnaire, and keep the coding consistent in SPSS.

What Cronbach's Alpha is acceptable for an attitude scale?

Alpha from 0.7 is commonly referenced according to (Nunnally, 1978). For a new scale or exploratory research, a level from 0.6 may be considered according to (Hair et al., 2010). You still need to examine Corrected Item-Total Correlation, the item content, and the reason for deleting any variable. Do not make the decision from Alpha alone.

Can I write the entire attitude scale myself?

You can develop new observed variables when no suitable scale exists for the construct or context, but you need to explain the basis for each statement and test the scale rigorously. For a student thesis, adopting and adapting an existing source is usually easier to defend, provided that you do not copy it mechanically.

What should I do if an attitude scale loses a variable?

First, check data-entry errors, reverse coding, unclear wording, and the fit between the variable and the construct. Then inspect the Item-Total Statistics or Rotated Component Matrix, record the reason for deletion, and rerun the procedure. If deleting the variable removes an important aspect, discuss it with your supervisor instead of keeping only the result with the most attractive statistic.

Open your questionnaire file and .sav or .csv file again. Create a separate table for the variable code, source, coding method, and decision to retain or delete each variable. When you need help running the analysis on your own SPSS data, see M4, DoThesis data analysis.