What Is Reliability? How to Read and Assess Scale Reliability

What is reliability

You open the results table and see Cronbach's Alpha, Composite Reliability, rho_A, or McDonald's Omega. They all contain the word reliability, but they are not the same statistic. In a quantitative thesis, reliability usually means scale reliability, or the consistency among the items that measure the same concept.

For example, a Satisfaction scale may contain SAT1, SAT2, SAT3, and SAT4. If a respondent tends to choose a high score for SAT1, the other items should show a reasonably consistent pattern, provided that they all reflect Satisfaction. Reliability does not tell you whether the scale measures the correct concept. It tells you whether the items operate consistently.

With Cronbach's Alpha, the software estimates reliability from the number of items and the correlations among them. The formula is commonly presented as:

α = k/(k - 1) × [1 - sum of the variance of each item / variance of the total score]

Here, k is the number of items. In a PLS-SEM model, Composite Reliability, commonly written as CR, is calculated from the outer loadings of the items. You therefore need to distinguish the reliability of a scale from the reliability of the entire research model. For more background on concepts such as scale, see what is a scale.

Why reliability matters in quantitative research

Reliability is a check you make before interpreting relationships among variables. If the items in one scale are inconsistent, the score representing the latent variable may contain substantial noise. EFA, regression, SEM, or PLS-SEM results then become harder to explain.

In a common workflow, you check reliability after cleaning the data and before exploratory factor analysis. In SPSS, you can run Cronbach's Alpha for each group of items. In SmartPLS, you review outer loading, Cronbach's Alpha, rho_A, and Composite Reliability together in the measurement model assessment.

Reliability answers this question: are the items in a scale consistent enough to retain and use in the next step. It does not answer questions about causal effects, p-values, R², or whether a hypothesis is significant. For example, a scale can have good CR while its HTMT result fails, or the path between two latent variables can be statistically insignificant.

You should also separate reliability from validity. Reliability concerns consistency. Convergent validity asks whether the items converge on the same concept, usually based on outer loading and AVE. Discriminant validity asks whether different concepts overlap too much, and is commonly assessed with HTMT or the Fornell-Larcker criterion.

If your research includes a mediator or moderator, check reliability for each construct before analysing mediation or moderation. You can review these concepts in mediator, moderator, and moderating, but do not use reliability to conclude that a variable is a mediator or moderator.

What reliability value is acceptable

Do not apply one number mechanically to every study. The threshold depends on the type of reliability, the research purpose, the number of items, and whether the scale is new or already established. The table below summarises commonly used thresholds and gives you a source to cite in Chapter 3 or Chapter 4.

These thresholds are reference points, not automatic reasons to delete an item. For example, a very high Alpha can occur because the items repeat the same idea too closely. A low Alpha can result from having few items, failing to recode a reverse-coded item, or using items that do not actually measure the same concept.

CR is an indicator you will often see in SmartPLS and AMOS. Its calculation and interpretation are not exactly the same as running Cronbach's Alpha in SPSS. If you use CB-SEM with AMOS, report model-fit indicators according to the CB-SEM procedure. CFI, TLI, RMSEA, and SRMR from (Hu and Bentler, 1999) should not be inserted into an explanation of SmartPLS results as though they were PLS-SEM criteria.

How to read reliability output

Reading Cronbach's Alpha in SPSS

In SPSS, go to Analyze > Scale > Reliability Analysis. Move the items for one construct into the Items box, select Model: Alpha, and then click Statistics. Select Item, Scale, Scale if item deleted, and Correlations if you want more information for deciding whether to remove an item.

The Reliability Statistics table usually contains two important columns, Cronbach's Alpha and N of Items. The Item-Total Statistics table then shows Corrected Item-Total Correlation and Cronbach's Alpha if Item Deleted. Read both tables, because an acceptable overall Alpha does not mean that every item fits the scale.

Illustrative output

The table below is illustrative output, not the result of a real study. Assume that the Satisfaction scale has four items and that you ran the menu sequence above.

In this example, SAT3 is the item that requires checking, not an item that must definitely be deleted. Review its wording, reverse coding, missing data, and correlation with the other items. If SAT3 is reverse-coded and you did not recode it, the low result may come from a data error rather than from the scale itself.

Reading Composite Reliability in SmartPLS

In SmartPLS 4, after running PLS Algorithm, open the measurement model results and find the Construct Reliability and Validity table. The common columns include Cronbach's alpha, rho_A, Composite reliability (rho_C), and Average variance extracted (AVE).

If the table has a row named Satisfaction, read the CR for that row, not the CR for SAT1, SAT2, or SAT3. Reliability belongs to the construct, while outer loading belongs to each item. When you need to check which item is weak, open the Outer Loadings table and match the item codes.

Illustrative example: Satisfaction has Cronbach's Alpha of 0.842, rho_A of 0.849, Composite Reliability of 0.887, and AVE of 0.663. CR is above 0.70 and AVE is above 0.50, so the scale shows acceptable evidence of reliability and convergent validity according to (Fornell and Larcker, 1981). These are still illustrative output values. Replace them with the output from your own .sav or .csv file.

How to write this in your thesis

You can adapt the following sentence to your actual output: “The results show that the [CONSTRUCT NAME] scale has a Cronbach's Alpha of [VALUE], while the items have Corrected Item-Total Correlation values ranging from [LOWEST VALUE] to [HIGHEST VALUE]. Because these values meet the [THRESHOLD] according to [SOURCE], the scale was retained for the next stage of analysis.”

For SmartPLS, you can write: “Composite Reliability for the constructs ranges from [LOWEST VALUE] to [HIGHEST VALUE], and all values are [higher than or lower than] 0.70. The results were compared with the criterion of (Fornell and Larcker, 1981) and considered together with outer loading, AVE, and HTMT.” Use the phrase “all values meet the criterion” only when every row actually meets it.

What to do when reliability fails

Start by identifying which indicator fails. A low Alpha in SPSS requires a different response from a low CR in SmartPLS. Do not delete an item simply because the software displays a higher Alpha if Item Deleted value.

First, check the coding. If the questionnaire contains an item with a negative meaning, recode it before running reliability. In SPSS, use Transform > Recode into Different Variables, create a new variable, and document the coding rule. Avoid overwriting the original data before saving a backup.

Second, check missing data, values outside the Likert range, and unusual response patterns. A response sheet with the same value for every item can reduce data quality. You can read more about outliers, but an outlier is not automatically a response that you may delete without justification.

Third, review the content and Corrected Item-Total Correlation for each item. If one item is clearly low, respondents may have understood its wording differently from what you intended, or the item may not fit the context. Record the reason for deleting an item and the statistics before and after deletion. Do not run the analysis repeatedly until the table looks good and report only the final round.

Fourth, if the scale still fails, review the conceptual structure and the scale source. A construct may combine dimensions that should have been separated. Alpha can also be unstable when a scale has only two items. At that point, discussing the theoretical basis and measurement design with your supervisor is more useful than deleting additional items.

A reviewer will usually accept item deletion when you state the criterion clearly, provide a content-based reason, and report the procedure transparently. Reviewers are less likely to accept deleting many items simply to reach 0.70, or retaining an item with a low indicator without explanation.

Distinguishing reliability from similar indicators

Cronbach's Alpha is one way to assess internal consistency. Composite Reliability also assesses the reliability of a construct, but it is calculated from outer loading in the measurement model. You can report both indicators, but do not describe them as two identical measurements.

Corrected Item-Total Correlation is the correlation between one item and the total scale score after that item has been removed from the total. It helps you identify an item that does not move with the rest of the scale. This is an item-level indicator, while Alpha and CR are indicators at the scale or construct level.

AVE is Average Variance Extracted, and it shows how much average variance a construct explains in its items. AVE of 0.50 or above is used as a criterion for convergent validity according to (Fornell and Larcker, 1981). AVE therefore does not replace CR.

HTMT is an indicator of discriminant validity between constructs. A low HTMT does not prove high reliability, just as a high CR does not prove that the constructs are sufficiently distinct. Read each group of indicators according to the question it answers.

Common mistakes

The first mistake is using Alpha for the entire questionnaire to conclude that every scale is reliable. Run and report the analysis for each construct, such as Perceived Usefulness, Ease of Use, and Intention to Use.

The second mistake is seeing a high Alpha if Item Deleted value and immediately deleting the item. Check the item content, coding, distribution, and theoretical basis. A small increase in Alpha is not enough to justify removing content that is important to the construct.

The third mistake is mixing SPSS criteria with SmartPLS criteria. SPSS Reliability Statistics does not contain CR. In SmartPLS, the Construct Reliability and Validity table contains several columns, and each one must be interpreted using its exact name.

The fourth mistake is using CFI, TLI, and RMSEA to assess a SmartPLS model. These are CB-SEM fit indices discussed in (Hu and Bentler, 1999). If you run SmartPLS, focus on appropriate indicators such as outer loading, CR, AVE, HTMT, VIF, R², f², and Q².

The final mistake is reporting values that are rounded too heavily or inserting illustrative output into your thesis. Before submitting, reopen the original output and check the construct names, number of items, and decimal places. Every number in Chapter 4 must be traceable to your analysis file.

Frequently asked questions

What is reliability in SPSS?

In SPSS, reliability is commonly checked with Cronbach's Alpha through Analyze > Scale > Reliability Analysis. Read the Reliability Statistics table for the overall Alpha and the Item-Total Statistics table for the Corrected Item-Total Correlation of each item.

What CR value is acceptable for Composite Reliability?

CR of 0.70 or above is commonly used as a reference threshold for composite reliability according to (Fornell and Larcker, 1981). However, read CR together with outer loading, AVE, and HTMT rather than drawing a conclusion from one number alone.

Are Cronbach's Alpha and Composite Reliability the same?

Both indicators concern scale reliability, but their calculations and contexts differ. Alpha commonly appears in SPSS, while CR is usually reported in the measurement model assessment of SmartPLS or SEM.

What should I do if reliability is low?

Check the items in this order: coding, missing data, unusual values, Corrected Item-Total Correlation, and item content. Delete an item only when you have both statistical and theoretical support, and record each round of processing so you can explain it.

Do I need to report Alpha, rho_A, and CR?

If you use PLS-SEM, report the reliability indicators that fit your procedure and interpret them together. Do not add indicators simply to make the table longer. State the software, version, construct, value, and comparison criterion clearly.

Open your actual output now, prepare one table for each construct, and compare Alpha or CR with the appropriate criterion before writing the interpretation. If you need to run reliability on a .sav or .csv file and turn the results into a Chapter 4 paragraph, DoThesis M4 data analysis handles this step.