What Is Reliability? How to Read Cronbach's Alpha and Composite Reliability

What Is Reliability

You open the SPSS output and see Cronbach's Alpha = 0.812, but you are still unsure whether the scale is good enough to keep or whether an item should be removed. In quantitative research, reliability tells you whether the items in the same scale are consistent when they measure the same concept. A satisfaction scale, for example, may include items such as SAT1, SAT2, SAT3, and SAT4. Reliability checks whether these items work closely enough together to be treated as one group.

The most familiar index in SPSS is Cronbach's Alpha. In principle, Alpha depends on the number of items and the correlations among the items in the same scale. It can be represented in general terms as follows:

α = [k/(k - 1)] × [1 - total variance of the items / total scale variance]

Here, k is the number of items. A higher Alpha usually indicates better internal consistency, but a very high Alpha also needs attention because the items may repeat the same content. Reliability does not prove that a scale measures the intended concept correctly. You still need to assess convergent validity, discriminant validity, and model fit.

In SmartPLS, you will usually see Composite Reliability, rho_A, and Cronbach's Alpha in the measurement model assessment. Composite Reliability uses the outer loading of each item, so it fits the PLS-SEM context more closely when you assess a latent variable. You can review the basic concept of what a scale is before reading these indices.

Why Reliability Matters in Quantitative Research

Reliability is checked before you use a scale for EFA, regression, CFA, or a PLS-SEM model. If the items in a group are inconsistent, the group's mean or composite score may not represent the concept you want to measure well.

In SPSS, reliability testing usually helps you answer three questions. First, does the full scale have acceptable internal consistency? Second, which item has a low correlation with the rest of the scale? Third, would Alpha increase substantially if one item were removed?

When Alpha reaches an appropriate level and the Corrected Item-Total Correlation for each item meets the requirement, you have a basis for keeping the scale for the next step. The decision to keep or remove an item must also consider the theoretical content. An item with a low index may measure an important aspect, while an item that increases Alpha but duplicates another item may still be unsuitable to keep.

In SmartPLS, reliability belongs to the measurement model assessment. You usually examine it together with Outer Loadings, AVE, HTMT, and VIF. A sound process is to assess the measurement model first and then interpret Path Coefficients, R², or mediation effects. Concepts such as mediator and moderator belong to the structural model and do not replace the reliability check.

What Reliability Level Is Acceptable

You should not look at one Alpha value and make a conclusion immediately. Identify the index, the purpose of the study, and the software you are using. The table below summarises commonly used reference thresholds and gives you a source to cite in your thesis.

These thresholds are reference points, not automatic deletion rules. For a new scale, an Alpha from 0.6 may be considered in an appropriate context, but you need to state clearly that the study is exploratory and cite (Hair et al., 2010). For a scale that has been used consistently, your supervisor may require a stricter level.

A common mistake is to call Composite Reliability the SmartPLS version of Cronbach's Alpha. Both indices concern reliability, but their calculations and roles in the model differ. You should also not use CFI, TLI, or RMSEA to assess reliability. These are fit indices for CB-SEM and are reported in a different context according to (Hu and Bentler, 1999).

How to Read Reliability in the Output

Reading the Reliability Statistics Table in SPSS

To run the test, go to Analyze > Scale > Reliability Analysis. Move the items for one construct into Items, select Model: Alpha, and click Statistics. Select Item, Scale, and Scale if item deleted so that SPSS produces the additional information you need.

The Reliability Statistics table usually contains two columns, Cronbach's Alpha and N of Items. N of Items shows the number of items included in that run. If you run each construct separately, check that the correct group of items is selected before interpreting Alpha. Do not combine all independent and dependent variables in one run because they measure different concepts.

The table below is illustrative output, not the result of a specific study.

In the actual output, you will also see the Item-Total Statistics table. The columns to examine are Corrected Item-Total Correlation and Cronbach's Alpha if Item Deleted. The example below follows the way SPSS presents the results.

This is also illustrative output. All Corrected Item-Total Correlation values are above 0.3, which matches the threshold from (Nunnally and Bernstein, 1994). Alpha if Item Deleted is lower than the overall Alpha of 0.812 for every item, so this table gives no clear statistical reason to remove an item.

If Cronbach's Alpha if Item Deleted is higher than the overall Alpha, inspect the corresponding item. Do not delete it immediately. Check the item wording, its coding direction, and its correlations with the other items. A reverse-coded item that has not been recoded can sharply reduce the correlation.

Reading the Indices in SmartPLS

In SmartPLS 4, after running PLS-SEM Algorithm, open the construct results to view Outer Loadings, Cronbach's Alpha, rho_A, Composite Reliability, and AVE. Menu names may change slightly between versions, but the logic remains the same: run the algorithm first, then read the measurement model results.

The table above is illustrative output, not a finding from your data. Composite Reliability from 0.7 and AVE from 0.5 are commonly used reference points according to (Fornell and Larcker, 1981). You also need to check the outer loading of each item and HTMT between constructs. SmartPLS does not use the CFI, TLI, or RMSEA indices from CB-SEM as substitutes for this group of indices.

What to Do When Reliability Is Not Acceptable

Check Data Entry Errors and Reverse-Coded Items

Before removing an item, check missing values, Likert-scale coding, and variable names. If an item has a negative meaning but has not been recoded, its correlation with the other items may be low. In SPSS, use Transform > Recode into Different Variables to create a reverse-coded variable, then run the analysis again using the new variable.

You should also check unusual response rows. A respondent who selects the same value for every item may distort the results, but do not delete the case simply because the answers are identical. Set your data-cleaning criteria in advance and record the reason for removing each case. If the data contain extreme values, consult the guidance on outliers instead of changing values to increase Alpha.

Examine Corrected Item-Total Correlation

If an item has a Corrected Item-Total Correlation below 0.3, review its content and coding. This threshold is based on (Nunnally and Bernstein, 1994). When the item is genuinely unsuitable in terms of content and removing it improves the scale, you can remove it and run the analysis again.

Compare Alpha if Item Deleted

If removing an item increases Alpha slightly, that increase is not enough by itself to make the decision. Consider the number of remaining items, the content of the construct, and the requirements of the model. Repeatedly deleting items until Alpha is as high as possible can leave you with items that are almost duplicates.

Report Each Processing Round Honestly

EFA and Cronbach's Alpha rarely look clean on the first run. You may remove an item after checking both the statistics and the theoretical basis, but record which item was removed, why it was removed, and how Alpha changed. The committee will often ask about the logic behind item removal, not only the final number.

If Alpha is low because the scale covers several different aspects, combining all items into one construct may be the cause. Return to the theoretical model and examine whether the scale structure makes more sense. Deleting items in bulk is usually less defensible. If the data are not of sufficient quality, software cannot turn them into reliable evidence.

Distinguishing Reliability from Scale Validity

Reliability asks whether the items are consistent with one another. Scale validity asks whether the scale actually measures the intended concept. A scale can have a high Alpha and still lack discriminant validity if two constructs measure almost the same content.

Cronbach's Alpha is usually read in SPSS together with Corrected Item-Total Correlation. Composite Reliability and rho_A are usually read in SmartPLS. AVE reflects convergent validity, while HTMT and the Fornell-Larcker criterion are used to assess discriminant validity. Therefore, an acceptable Alpha does not mean that the entire measurement model is acceptable.

You also need to distinguish reliability from other analyses. Moderating tests whether one variable changes the relationship between two other variables. Outliers focuses on unusual observations. These analyses may affect data quality, but they are not scale reliability indices.

Common Mistakes

The first mistake is running Cronbach's Alpha for the entire questionnaire in one run. Each construct needs to be checked separately. For example, service quality, trust, and repurchase intention should not be combined just because they all use a Likert scale.

The second mistake is reporting only Alpha and ignoring Corrected Item-Total Correlation. The reader needs to know which item was removed and why. If you use SmartPLS, you should also report Composite Reliability, AVE, and outer loading according to the measurement model structure.

The third mistake is deleting items only to reach 0.9. An excessively high Alpha may indicate that the items repeat the same content. Place the statistical decision alongside the theoretical decision instead of optimising one number.

The fourth mistake is taking a threshold from an online article without citing a source in the thesis. In the criteria table, state the source clearly, such as (Nunnally, 1978), (Hair et al., 2010), or (Fornell and Larcker, 1981). If the study uses PLS-SEM, do not mix CB-SEM fit criteria into the SmartPLS section.

Another mistake is writing the results before checking the data file. Save the output, variable coding, number of processing rounds, and final tables. If your lecturer asks why SAT3 was kept, you should be able to open the table and explain the decision using your own data.

Frequently asked questions

What is reliability in SPSS?

Reliability in SPSS is the internal consistency among items belonging to the same scale. The usual index is Cronbach's Alpha, read in the Reliability Statistics table together with Corrected Item-Total Correlation in the Item-Total Statistics table.

What Cronbach's Alpha is acceptable?

An Alpha of 0.7 or above is commonly considered acceptable according to (Nunnally, 1978). For a new scale or exploratory research, an Alpha from 0.6 may be considered according to (Hair et al., 2010), but you need to state the context and avoid applying the number mechanically.

Does a low Alpha mean that I should delete an item immediately?

Check the coding, reverse-coded items, missing values, and scale content first. Then examine Corrected Item-Total Correlation and Cronbach's Alpha if Item Deleted. Remove an item only when both the statistics and the theoretical basis support the decision.

What Composite Reliability is acceptable in SmartPLS?

Composite Reliability from 0.7 is commonly used as a reference point according to (Fornell and Larcker, 1981). You still need to read it together with outer loading, AVE, and HTMT because Composite Reliability reflects only one part of measurement model quality.

Are Cronbach's Alpha and Composite Reliability the same?

Both indices concern reliability, but their calculations are different. Cronbach's Alpha is commonly used in SPSS, while Composite Reliability is widely reported in PLS-SEM. Do not replace one index with the other without explaining the research method.

How should I write reliability results in my thesis?

You can adapt the following sentence to your actual data: “The reliability test results show that the [SCALE NAME] scale has a Cronbach's Alpha of [ALPHA], with item Corrected Item-Total Correlation values ranging from [LOWEST VALUE] to [HIGHEST VALUE]. Therefore, [RETAIN/REMOVE] items [ITEM CODES] before conducting [EFA/FURTHER ANALYSIS].” Replace every bracketed section with the output from your analysis and state the source of the threshold used.

How to write this in your thesis

Open your file, run each scale separately using Analyze > Scale > Reliability Analysis, save both Reliability Statistics and Item-Total Statistics, and then compare them with the measurement model if you use SmartPLS. If you need to run the analysis on your own .sav or .csv file and check each processing round, you can use M4 data analysis.